Insight of the Week: Work faster and smarter with ChatGPT
If your boss hasn't already asked you how you're leveraging ChatGPT (and other Large Language Models, LLMs) I'd wager that they will soon.
Progress in this particular field of AI has been incredible over the last several years. And especially the last few months, since the launch of ChatGPT in November last year.
In an excellent?Stanford Online webinar, Professor Christopher Potts showed a compelling illustration of this:
The orange horizontal line is the level at which humans perform. The other lines plot historical progress of AI towards (and beyond) that level of performance for certain tasks, known as benchmarks. New benchmarks get created every now and again, primarily to test AI in different, more challenging scenarios.
Don't sweat the details... just notice how vertical the lines are becoming. It took five or six years for AI to exceed human capabilities on benchmarks created in the late nineties. In more recent benchmarks, AI beat humans within 12 months.
I think this tells us that the tools and techniques of AI are becoming more flexible and robust. We're still a way away from what's know as Artificial General Intelligence (AGI) where AI achieves, or exceeds, the flexibility and robustness of human intelligence.
But there's no question that AI got a lot more useful in the last 6 months. So back to the question that your boss will be asking soon, if not already: "how are you leveraging LLMs in your business?"
I like to think about it on three levels: 1) Helping us do our jobs better, 2) Helping existing AI tech work better and 3) Using LLMs to dynamically generate responses to customer queries
I'll cover the first of these here, and cover areas 2 and 3 in future emails.
So, how can LLMs help us do our jobs better?
In short, LLMs promise to help you work faster, and smarter. Microsoft are soon going to be rolling these capabilities into their entire office suite, and the demos look compelling. In the meantime, you can use the AI integrations in digital note-taking tools like?notion.so?or go straight to?chat.openai.com?and use Chat GPT.
Here are some ideas:
Summarize:?Provide a long-form text and ask for a short summary. Example: Summarize this in 50 words [full text of the US Declaration of Independence]. Here's what Chat GPT came up with:?
领英推荐
On July 4, 1776, the thirteen united states of America declared their independence from Great Britain with the unanimous Declaration. The document asserts that all men are created equal, have unalienable rights, and are entitled to life, liberty, and the pursuit of happiness. It states that the purpose of government is to secure these rights, and when it becomes destructive to them, the people have the right to alter or abolish it. The Declaration also lists grievances against King George III, declaring him a tyrant who has violated the colonists' rights and freedoms.
Research:?Ask a question. Be specific about how you want it answered. Example: Provide a 50 word summary on the advantages and disadvantages of LLMs. Here's what Chat GPT came up with:
Large Language Models (LLMs) such as GPT-3 have the advantage of being able to perform a wide range of language-related tasks with high accuracy and can generate human-like text. However, they are expensive to train, require large amounts of data, and can perpetuate biases and inaccuracies in language models if not trained and tested carefully.
Reword:?Take something complicated and have an LLM communicate it in a different way. E.g. I gave ChatGPT the first paragraph of the US constitution and asked it to explain it to a ten-year old it responded:
This sentence is the beginning of the United States Constitution. It says that the people of the United States got together to create a document that would help make their country better. They wanted to make sure that everyone was treated fairly and that there was peace and safety in their country. They also wanted to make sure that everyone could live a good life and be free. So they wrote down some rules for how the country should be run, and that's what this document is.
Coding and Code explanation:?This is a whole subject in of itself, but know that LLMs are getting really good at creating and explaining code. I asked ChatGPT to explain a complex Excel formula. It did a really good job, but it was quite a long explanation, so here’s an abbreviated version of the response:
The `INDEX-MATCH` formula in Excel is a combination of two different functions that work together to retrieve a value from a specific cell in a table… the formula is looking for the value in a table located in cells C3:E9, where the row of the desired value matches the position of the first lookup value in cells C3:C9, and the column of the desired value matches the position of the second lookup value in cells C3:E3.
LLMs can also help generate test cases for quality assurance and create first drafts of documents, among many, many other use cases.?
But the way you ask the question is important. Be specific, and if you want to go deeper, you need to start learning about ‘prompt engineering’. A good place to start is to go to?chat.openai.com?and ask ChatGPT:?List 3 examples of prompt engineering???
Just remember, LLMs are prone to 'hallucinations' - where they say things that sound plausible, but are completely wrong. Use the LLM to work faster and smarter, but you remain responsible for how you use the results!
My team and others at Waterfield are already leveraging LLMs to research industries, generate draft designs and reports, and to create test cases, among many others. In next week’s email, I’ll look at how LLMs can help us get more out of our existing AI platforms.
How are you using LLMs to do your job better?
Kerry Robinson?is an Oxford physicist with a Master's in Artificial Intelligence. Kerry is a technologist, scientist, and lover of data with over 20 years of experience in conversational AI. He combines business, customer experience, and technical expertise to deliver IVR, voice, and chatbot strategy and keep Waterfield Tech buzzing.